[cpp.line] Simplify specification of line directives

[AlisdairM]

This change set addresses a variety of concerns to produce a simpler and clearer specification for line directives. It makes no functional changes, but changes words sufficiently to require a Core review. A motivation behind this PR is to simplify the changes for the proposal to allow line directives before the global module fragment, https://wg21.link/P3868.

The line directive is given its own grammar production that formally states what used to be described in text restrictions and interpretations of the generic sequence of preprocessing tokens.

The other issue to address is that the specification refers to a string-literal that is not present in the grammar. However, the constraints on that string literal are sufficient to replace that quoted s-char-sequence in the grammar production with string-literal.

[jensmaurer]

Remark (not necessarily dispositive): The status quo still has a good match with the C phrasing; this change would kill that correspondence.

Other than that, the funny presentation of the preprocessor grammar in the status quo has at least two useful properties that we would lose with the reformulation:

• There is no question that # line pp-tokens is a control-line. Once you factor the "of the form" rules into the grammar, the parser would simply reject a parse as a control-line and instead attempt to treat that as a text-line. Which is not desirable.
• As written in the pull request, the grammar seems woefully ambiguous, because #line 100 matches both of the #line alternatives (100 is a preprocessing-token, indeed).

[AlisdairM]

The grammar is disambiguated the same way that we do for #include and #embed but referring to "the first form" and going to the second form only if the first does not match.

The control-line issue is a genuine issue that I missed and should consider more carefully. That should also be a concern for #8249.